| In recent years,with the continuous improvement of ship automation level in the field of shipping,the digitization and intelligent broadcast of safety information are also improving.And NAVTEX message is an important part of it.However,the current problem is that these NAVTEX messages are difficult to be efficiently applied to ship driving management and decision support.For example,after the seafarers receive new NAVTEX messages,they have to be manually read to classify them and then analyzed to make the corresponding navigation decisions.Not only could this involve low work efficiency,but also a greater chance of error in a process with human involvement compared to an automated process.To address this issue,most of the current research on NAVTEX has been conducted to improve the performance of its system in terms of communication technology.The research on semantic classification of its message contents and providing contextual decision-making services for seafarers is not mature enough,and there is plenty of space for expansion.Therefore,this paper constructs a NAVTEX message semantic classification and contextual decision support system based on the technology of NLP.Moreover,in the face of the problem of message contextual classification in the system,ML-BLS is proposed by improving the BLS.The classification effect of the system is more accurate by using this model to solve the multi label classification problem.And the effectiveness of the model is verified in the experiment.The decision support system consists of three parts: data preprocessing module,feature extraction module and classification model building module.In the process of building the system,the following studies are conducted:(1)In the data preprocessing module,two sets of labels,semantic label and contextual label,are proposed to label the categories of messages;Among them,the semantic labels and messages are in a one-to-many relationship,which is a part of the single-label classification;while the contextual labels and messages are in a many-to-many relationship,which is a part of the multi-label classification.(2)For the feature extraction module,the VSM,LDA and Doc2 vec methods are combined with SVM,ANN and BLS single-label classification models for comparative experiments,and with ML-NN and ML-BLS multi-label classification models for comparative experiments.The results show that the performance of VSM is better than that of LDA and Doc2 vec.It is more suitable for feature extraction of message.Finally,the optimal parameters of the feature vector dimension of VSM are determined by experiments to be 300 for semantic classification and 200 for contextual classification.(3)For the classification model building module,the above comparison experiments show that BLS and ML-BLS work best for semantic and contextual classification,respectively.Through experiments to determine the best parameters of BLS and ML-BLS,the accuracy of BLS can reach 99.92%,and the Micro-F1 of MLBLS can reach 99.96%.And the classification effect of each class of semantic labels is studied to verify the effectiveness of combining VSM and BLS models in semantic classification.At the end of this paper,a specific application case study is conducted to verify the effectiveness of this system,and the specific process is as follows: Firstly,after the data preprocessing module completes the message import,it carries out data cleaning and word segmentation;Secondly,the preprocessed messages are vectorized and dimensionality reduced in the feature extraction module;Then,the vectorized messages are input into the classification model building module for semantic and contextual classification to get the results,and the decision support during navigation is completed.Finally,the application analysis of the system is carried out.Through the above research work in this paper,information,automation and intelligent technology are used in the field of shipping,which not only expands the application scope of NLP,but also provides intelligent decision support services for seafarers.It can reduce the workload of seafarers and reduce their work intensity.For example,it is no longer necessary to manually read the messages received during navigation for semantic classification and contextual decision making,so that further enhance the safe navigation performance of the ship.In addition,it is faster and more convenient to search for messages after they are archived.The NAVTEX message semantic classification and contextual decision support system based on NLP constructed in this paper is expected to be further integrated into intelligent ships in the future extended research,to enrich the functions of navigational aids. |
